Skip to content
master
Switch branches/tags
Go to file
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.

README.md

Learning Extremely Shared Middle-level Image Representation for Scene Classification

Introduction

The implemention codes on MIT-indoor 67 dataset.

A novel method to learn very compact and discriminative image representation.

Dependencies

The MIT-indoor 67 dataset can be downloaded at this website and should be put under the "data" path.

The caffe is needed to be installed.

The CNN model can be downloaded at the caffe model zoo and be put under the "feature_extraction_imagenet" path.

VLFeat and LIBLINEAR are also needed.

Trained models

We put 4 different trained models "best_2.mat", "best_4.mat", "best_6.mat", "best_8.mat" under the "data" path, which correspond to 2 patterns, 4 patterns, 6 patterns, 8 patterns per class respectively.

Testing

Run the "MITindoor_classification.m" to test the trained model.

Training

Run the "exp_MITindoorft.m" to train the model.

License

The code is released under the MIT License.

Citing paper

If you find the work is useful in your research, please considering citing:

@article{tang2016learning,
    title = {Learning extremely shared middle-level image representation for scene classification},
    author = {Tang, Peng and Zhang, Jin and Wang, Xinggang and Feng, Bin and Roli, Fabio and Liu, Wenyu},
    journal = {Knowledge and Information Systems},
    pages = {1--22},
    year = {2016},
    publisher = {Springer}
}

About

Codes of the paper "Learning Extremely Shared Middle-level Image Representation for Scene Classification"

Resources

Releases

No releases published

Packages

No packages published